Principles of Human Cognition Utilized for Automated Image Analysis Gerd Binnig 3.Nov.2009
I think therefore I am
Context Driven Image Analysis Context driven through CNL Object oriented Knowledge based context ?? YES
Which Object ?
Which Object ?
Context
Context
Object Oriented and Knowledge Based Thinking We have abstract  knowledge  about   classes  of  objects  and their  relations Fork! Knife! Plate between
Context Objects –  From the Concrete to the Vague Objects 1. CONTEXT Objects 2. CONTEXT  Object
Context Navigation  Image Plate ? ? ? ? Fork Handle Knife Context navigation - most important Handle Blade +1 0 +1 0 0
Context Navigation  Context slowly builds up   The easy and well defined “objects” first Image Plate ? ? ? ? Fork Handle Knife Handle
Only vague objects ?
Network of Objects   The objects are undefined The relations are concrete
Cognition Network Language - CNL Context driven  Knowledge based Object based Local Script based
Process Hierarchy  programming by multiple choice   . Knowledge Hierarchy  natural formulation   . Navigation  for directed local processing   . Object Hierarchy evolves in the course of processing   The Elements of CNL MRS, Chessboard, Merge, Classification, .. Fork! Knife! Plate between
Context driven  Knowledge based Object based Object-Pixel unification Local Script based Multi-Image capability Up to 4 Dimensions The Elements of CNL – Version 8
Process Hierarchy  programming by multiple choice   . Knowledge Hierarchy  natural formulation   . Navigation  for directed local processing   . Object Hierarchy evolves in the course of processing   The Elements of CNL – V8 MRS, Chessboard, Merge, Classification, .. Region O   +   P
Real Space / Size of Objects   www.definiens.com www.definiens.com 1nm 1  m 1mm 10 100 10 100 0.1 100 1m 10 1Km Atom transistor Organ Person Ship Car House City Cell Forest m=meter   Traveling through the Dimensions of Space 100
Phase Contrast Mic. Cells 3D-Confocal Microscopy   Cell biology 3D-Confocal Microscopy  Tissue  Molecular Pathology 3D  PET/CT  Small animal Biological Research and Drug Development CT   Mouse High Content Sreening Cells Proliferation index Tissue Cancer Biomarker Tissue
Biopsy Tissue Clinical Applications Serum Cells X-Ray Mammography CT Organs MRI  low res. Organs CT Head/Neck MRI Ventricles CT Lymph Nodes
Context Driven Processing – Cell Cultures -1 +1 3. Micro-   Nucleus Context   Objects :  Nuclei Cells 1. Nuclei 2. Cells Image
Context   Objects  – Blue, and Red Areas; Nucleus, Cell Blue stain red stain Different stains    – different contrasts Blue area Red area -1 -1 Nucleus Cell Membrane +1 -1 Nuc. Cell Membrane +1 -1 Context Driven Processing – Tissue
Spinal Cord +1 Context Driven Processing – CT Spine Context   Objects :  Spinal cord Liver Kidney Spleen Spine
Solution: Anatomical Context Anatomy segmentation Context-free lymph node  segmentation produces many false positives Reduced false positive rate + =
 
Hannover – University – Detecting HOUSES
Screenshots from Internet (Google)
Hannover – University – Detecting HOUSES What is easy and well defined? ..and can be used as context?
Meadow Potential shadow of tree 0 Context Driven Processing – Earth Observation 0 tree Direction = -55 ° Image 1. 2. 3. 4. 5.
Context Driven Processing – Earth Observation
Segmentation Result from Street-Model (also centerline)
Segmentation Result from Image –  Houses, Trees, Roads and Meadows
Centerlines of Roads Imprinted into Segmentation Result
Conflicts – Streets Run Through Houses
Finding Conflicts Automatically – in Blue
Demo
Different kind of context object
8:35  Umtata - South Africa 7604 x 4660 pixels
8:35  Multi-Resolution Segmentation
10:00  + Merge
4:30  Multi-Resolution Segmentation on Half Resolution
3:45  Pixel-Based
Major Roads Within and Outside the City (South Africa)
Major Roads Within and Outside the City (South Africa)
Rural Road Network near Mvezo, South Africa
Munich  -  100 MPixels
32:00
34:00
16:00
3:01  Half Resolution
Different kind of context object
Local Processing
Local Processing
Local Processing
Yield in Case of Dependencies Context objects need to be very reliable Chance for success: 0.95*0.95*0.95*0.95*0.95*0.95 =  0.73 Probabilities multiply
 
 
 
From  GIS   to   GIN   (From geographic Information Systems to a gigantic Geographic Information Network)
From  GIS  to  GIN A network of organizations, individuals, and  autonomous machines
From GIS to GIN Data provider Government Institution University Satellite Service provider Airplane sensor sensors people people people Drone (Internet) sensor Lidar sensor sensor The Emerging Network Relatively new (in red): New sensors, internet service providers, contribution of private individuals networking of sensors and their data, sensors on people, automated data creation
GIN   with  Intelligent Processing and Autonomous Machines (IP and AM) Data provider Government Institution University Satellite Airplane sensor sensor sensor sensor people people people sensors Drone Service provider (Internet) IP AM warning Lidar IP IP, AM IP Distributed sensors AM Company, Organization
Autonomous Machines ( AM ) Automatic data creation Unmanned vehicles satellites, drones stationary cameras Automatic data analysis Definiens- nothing else
Statement: automated data analysis is possible Analysis gets better and better Analysis gets simpler and simpler Combination of different data
Statement: automated data analysis is possible Analysis gets better and better XD Analysis gets simpler and simpler Combination of different data
Statement: automated data analysis is possible Analysis gets better and better XD Analysis gets simpler and simpler Combination of different data
Example for Simplicity: Combination of Infrared and Lidar Infrared=vegetation
Digital Surface Model (DSM) Lidar: Elevation=vegetation or building
Classification of Aerial Image with DSM (Buildings and Vegetation)  Buildings=Elevation-Vegetation
Intelligent automation is possible Consequences? … not only for image analysis
From  GIS  to  GIN   (geographic information network) A network of organizations, individuals, and autonomous machines
More data New types of data More automated data generation More private use of geographical information Private contribution More networking of different data More networking of organizations and people The future context for automated image analysis
From GIS to GIN   (Geographic Information Network) Data provider Government Institution University Satellite Service provider Airplane sensor sensor people people people Drone (Internet) sensor Lidar sensor sensor
More data New types of data More automated data generation More private use of geographical information Private contribution More networking of different data More networking of organizations and people Automated intelligent data processing Automated data and event communication The future context for automated image analysis
GIN   Plus  Intelligent Machines (IM) Data provider Government Institution University Satellite Airplane sensor sensor sensor sensor people people people sensor Drone Service provider (Internet) IM IM warning Lidar
GIN   Plus  IM  Plus  Pervasive Computing Data provider Government Institution University Satellite Airplane sensor sensor sensor sensor people people people sensor Drone Lidar Service provider (Internet) IM warning IM
Explosion of data: intelligent data processing is a solution  More data are turned into information Easier access to relevant information Explosion of information: context driven data management required Consequence of the consequences

E Cognition User Summit2009 G Binnig Definiens

  • 1.
    Principles of HumanCognition Utilized for Automated Image Analysis Gerd Binnig 3.Nov.2009
  • 2.
  • 3.
    Context Driven ImageAnalysis Context driven through CNL Object oriented Knowledge based context ?? YES
  • 4.
  • 5.
  • 6.
  • 7.
  • 8.
    Object Oriented andKnowledge Based Thinking We have abstract knowledge about classes of objects and their relations Fork! Knife! Plate between
  • 9.
    Context Objects – From the Concrete to the Vague Objects 1. CONTEXT Objects 2. CONTEXT Object
  • 10.
    Context Navigation Image Plate ? ? ? ? Fork Handle Knife Context navigation - most important Handle Blade +1 0 +1 0 0
  • 11.
    Context Navigation Context slowly builds up The easy and well defined “objects” first Image Plate ? ? ? ? Fork Handle Knife Handle
  • 12.
  • 13.
    Network of Objects The objects are undefined The relations are concrete
  • 14.
    Cognition Network Language- CNL Context driven Knowledge based Object based Local Script based
  • 15.
    Process Hierarchy programming by multiple choice . Knowledge Hierarchy natural formulation . Navigation for directed local processing . Object Hierarchy evolves in the course of processing The Elements of CNL MRS, Chessboard, Merge, Classification, .. Fork! Knife! Plate between
  • 16.
    Context driven Knowledge based Object based Object-Pixel unification Local Script based Multi-Image capability Up to 4 Dimensions The Elements of CNL – Version 8
  • 17.
    Process Hierarchy programming by multiple choice . Knowledge Hierarchy natural formulation . Navigation for directed local processing . Object Hierarchy evolves in the course of processing The Elements of CNL – V8 MRS, Chessboard, Merge, Classification, .. Region O + P
  • 18.
    Real Space /Size of Objects www.definiens.com www.definiens.com 1nm 1  m 1mm 10 100 10 100 0.1 100 1m 10 1Km Atom transistor Organ Person Ship Car House City Cell Forest m=meter Traveling through the Dimensions of Space 100
  • 19.
    Phase Contrast Mic.Cells 3D-Confocal Microscopy Cell biology 3D-Confocal Microscopy Tissue Molecular Pathology 3D PET/CT Small animal Biological Research and Drug Development CT Mouse High Content Sreening Cells Proliferation index Tissue Cancer Biomarker Tissue
  • 20.
    Biopsy Tissue ClinicalApplications Serum Cells X-Ray Mammography CT Organs MRI low res. Organs CT Head/Neck MRI Ventricles CT Lymph Nodes
  • 21.
    Context Driven Processing– Cell Cultures -1 +1 3. Micro- Nucleus Context Objects : Nuclei Cells 1. Nuclei 2. Cells Image
  • 22.
    Context Objects – Blue, and Red Areas; Nucleus, Cell Blue stain red stain Different stains – different contrasts Blue area Red area -1 -1 Nucleus Cell Membrane +1 -1 Nuc. Cell Membrane +1 -1 Context Driven Processing – Tissue
  • 23.
    Spinal Cord +1Context Driven Processing – CT Spine Context Objects : Spinal cord Liver Kidney Spleen Spine
  • 24.
    Solution: Anatomical ContextAnatomy segmentation Context-free lymph node segmentation produces many false positives Reduced false positive rate + =
  • 25.
  • 26.
    Hannover – University– Detecting HOUSES
  • 27.
  • 28.
    Hannover – University– Detecting HOUSES What is easy and well defined? ..and can be used as context?
  • 29.
    Meadow Potential shadowof tree 0 Context Driven Processing – Earth Observation 0 tree Direction = -55 ° Image 1. 2. 3. 4. 5.
  • 30.
    Context Driven Processing– Earth Observation
  • 31.
    Segmentation Result fromStreet-Model (also centerline)
  • 32.
    Segmentation Result fromImage – Houses, Trees, Roads and Meadows
  • 33.
    Centerlines of RoadsImprinted into Segmentation Result
  • 34.
    Conflicts – StreetsRun Through Houses
  • 35.
  • 36.
  • 37.
    Different kind ofcontext object
  • 38.
    8:35 Umtata- South Africa 7604 x 4660 pixels
  • 39.
  • 40.
    10:00 +Merge
  • 41.
    4:30 Multi-ResolutionSegmentation on Half Resolution
  • 42.
  • 43.
    Major Roads Withinand Outside the City (South Africa)
  • 44.
    Major Roads Withinand Outside the City (South Africa)
  • 45.
    Rural Road Networknear Mvezo, South Africa
  • 46.
    Munich - 100 MPixels
  • 47.
  • 48.
  • 49.
  • 50.
    3:01 HalfResolution
  • 51.
    Different kind ofcontext object
  • 52.
  • 53.
  • 54.
  • 55.
    Yield in Caseof Dependencies Context objects need to be very reliable Chance for success: 0.95*0.95*0.95*0.95*0.95*0.95 = 0.73 Probabilities multiply
  • 56.
  • 57.
  • 58.
  • 59.
    From GIS to GIN (From geographic Information Systems to a gigantic Geographic Information Network)
  • 60.
    From GIS to GIN A network of organizations, individuals, and autonomous machines
  • 61.
    From GIS toGIN Data provider Government Institution University Satellite Service provider Airplane sensor sensors people people people Drone (Internet) sensor Lidar sensor sensor The Emerging Network Relatively new (in red): New sensors, internet service providers, contribution of private individuals networking of sensors and their data, sensors on people, automated data creation
  • 62.
    GIN with Intelligent Processing and Autonomous Machines (IP and AM) Data provider Government Institution University Satellite Airplane sensor sensor sensor sensor people people people sensors Drone Service provider (Internet) IP AM warning Lidar IP IP, AM IP Distributed sensors AM Company, Organization
  • 63.
    Autonomous Machines (AM ) Automatic data creation Unmanned vehicles satellites, drones stationary cameras Automatic data analysis Definiens- nothing else
  • 64.
    Statement: automated dataanalysis is possible Analysis gets better and better Analysis gets simpler and simpler Combination of different data
  • 65.
    Statement: automated dataanalysis is possible Analysis gets better and better XD Analysis gets simpler and simpler Combination of different data
  • 66.
    Statement: automated dataanalysis is possible Analysis gets better and better XD Analysis gets simpler and simpler Combination of different data
  • 67.
    Example for Simplicity:Combination of Infrared and Lidar Infrared=vegetation
  • 68.
    Digital Surface Model(DSM) Lidar: Elevation=vegetation or building
  • 69.
    Classification of AerialImage with DSM (Buildings and Vegetation) Buildings=Elevation-Vegetation
  • 70.
    Intelligent automation ispossible Consequences? … not only for image analysis
  • 71.
    From GIS to GIN (geographic information network) A network of organizations, individuals, and autonomous machines
  • 72.
    More data Newtypes of data More automated data generation More private use of geographical information Private contribution More networking of different data More networking of organizations and people The future context for automated image analysis
  • 73.
    From GIS toGIN (Geographic Information Network) Data provider Government Institution University Satellite Service provider Airplane sensor sensor people people people Drone (Internet) sensor Lidar sensor sensor
  • 74.
    More data Newtypes of data More automated data generation More private use of geographical information Private contribution More networking of different data More networking of organizations and people Automated intelligent data processing Automated data and event communication The future context for automated image analysis
  • 75.
    GIN Plus Intelligent Machines (IM) Data provider Government Institution University Satellite Airplane sensor sensor sensor sensor people people people sensor Drone Service provider (Internet) IM IM warning Lidar
  • 76.
    GIN Plus IM Plus Pervasive Computing Data provider Government Institution University Satellite Airplane sensor sensor sensor sensor people people people sensor Drone Lidar Service provider (Internet) IM warning IM
  • 77.
    Explosion of data:intelligent data processing is a solution More data are turned into information Easier access to relevant information Explosion of information: context driven data management required Consequence of the consequences